NVIDIA 2D Thread Scheduler Fixed #421
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Related issue: #356 |
stratika
reviewed
May 14, 2024
...s/opencl/src/main/java/uk/ac/manchester/tornado/drivers/opencl/TornadoPlatformInterface.java
Outdated
Show resolved
Hide resolved
| } | ||
|
|
||
| private long[] calculateEffectiveMaxWorkItemSizes(TaskMetaData metaData) { | ||
| long[] intermediates = new long[] { 1, 1, 1 }; |
There was a problem hiding this comment.
any particular reasoning for this particular name of the array? Shouldn't this reflect the max work items?
There was a problem hiding this comment.
This is to calculate the local Work Group. We initialize it to one and then fill with the dimensions used.
There was a problem hiding this comment.
yes, but the name is intermediates? can we make it more specific?
There was a problem hiding this comment.
Yes, I agree. I just did a refactoring of this method
…drivers/opencl/TornadoPlatformInterface.java Co-authored-by: Thanos Stratikopoulos <34061419+stratika@users.noreply.github.com>
|
I guess for older driver versions, we will not see any difference, right? I tried it with |
…uler' into feat/nvidia/scheduler
mikepapadim
approved these changes
May 14, 2024
jjfumero
added a commit
to jjfumero/TornadoVM
that referenced
this pull request
May 28, 2024
Improvements ~~~~~~~~~~~~~~~~~~ - beehive-lab#402 <beehive-lab#402>: Support for TornadoNativeArrays from FFI buffers. - beehive-lab#403 <beehive-lab#403>: Clean-up and refactoring for the code analysis of the loop-interchange. - beehive-lab#405 <beehive-lab#405>: Disable Loop-Interchange for CPU offloading.. - beehive-lab#407 <beehive-lab#407>: Debugging OpenCL Kernels builds improved. - beehive-lab#410 <beehive-lab#410>: CPU block scheduler disabled by default and option to switch between different thread-schedulers added. - beehive-lab#418 <beehive-lab#418>: TornadoOptions and TornadoLogger improved. - beehive-lab#423 <beehive-lab#423>: MxM using ns instead of ms to report performance. - beehive-lab#425 <beehive-lab#425>: Vector types for ``Float<Width>`` and ``Int<Width>`` supported. - beehive-lab#429 <beehive-lab#429>: Documentation of the installation process updated and improved. - beehive-lab#432 <beehive-lab#432>: Support for SPIR-V code generation and dispatcher using the TornadoVM OpenCL runtime. Compatibility ~~~~~~~~~~~~~~~~~~ - beehive-lab#409 <beehive-lab#409>: Guidelines to build the documentation. - beehive-lab#411 <beehive-lab#411>: Windows installer improved. - beehive-lab#412 <beehive-lab#412>: Python installer improved to check download all Python dependencies before the main installer. - beehive-lab#413 <beehive-lab#413>: Improved documentation for installing all configurations of backends and OS. - beehive-lab#424 <beehive-lab#424>: Use Generic GPU Scheduler for some older NVIDIA Drivers for the OpenCL runtime. - beehive-lab#430 <beehive-lab#430>: Improved the installer by checking that the TornadoVM environment is loaded upfront. Bug Fixes ~~~~~~~~~~~~~~~~~~ - beehive-lab#400 <beehive-lab#400>: Fix batch computation when the global thread indexes are used to compute the outputs. - beehive-lab#414 <beehive-lab#414>: Recover Test-Field unit-tests using Panama types. - beehive-lab#415 <beehive-lab#415>: Check style errors fixed. - beehive-lab#416 <beehive-lab#416>: FPGA execution with multiple tasks in a task-graph fixed. - beehive-lab#417 <beehive-lab#417>: Lazy-copy out fixed for Java fields. - beehive-lab#420 <beehive-lab#420>: Fix Mandelbrot example. - beehive-lab#421 <beehive-lab#421>: OpenCL 2D thread-scheduler fixed for NVIDIA GPUs. - beehive-lab#422 <beehive-lab#422>: Compilation for NVIDIA Jetson Nano fixed. - beehive-lab#426 <beehive-lab#426>: Fix Logger for all backends. - beehive-lab#428 <beehive-lab#428>: Math cos/sin operations supported for vector types. - beehive-lab#431 <beehive-lab#431>: Jenkins files fixed.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
This patch provides a new Thread Scheduler for NVIDIA GPUs.
Problem description
The problem is that, when using the latest NVIDIA Drivers (e.g., 550.76), the thread block is set to 32x32 for 2D kernels. This block size seems to be illegal only when using the latest NVIDIA drivers. This patch provides a custom NVIDIA scheduler to fix this. Performance over the default scheduler increases ~300GFLOPs on my RTX 3070 GPU for the canonical matrix multiplications with this patch.
Backend/s tested
Mark the backends affected by this PR.
OS tested
Mark the OS where this PR is tested.
Did you check on FPGAs?
If it is applicable, check your changes on FPGAs.
How to test the new patch?